Annual models from DMED data.

Scope:

CONUS Army Installations:

Fort Jackson, SC     
Fort Benning, GA    
Fort Bragg, NC  
Fort Campbell, KY  
Fort Polk, LA  
Fort Hood, TX  
Fort Stewart, GA  
Fort Leonard Wood, MO
Fort Riley, KS  
Fort Irwin, CA  
Fort Bliss, TX  
  

Population: Active-duty Army service members

Outcome: Ambulatory (Out-patient)
Annual Rate of Heat Stress Illness (any type, primary diagnosis)
1997 - 2018

Exposure indices:
“Absolute” indices
Annual mean (full-year): temperature, heat index, WBGT
Annual heat risk days / hours - Heat index above 80 / 90 / 103 / 125 °F - WBGT above 82 / 85 / 88 / 90 °F

“Relative” indices (averaged over full-year and heat season months) Annual mean daily anomaly: temperature, heat index, WBGT
Annual maximum daily anomaly: temperature, heat index, WBGT
Days mean temperature index above daily climate normal percentile (averaged over all hours of day)
- temperature, heat index, WBGT above 85th / 90th / 95th percentiles
Days maximum temperature index above daily climate normal maximum percentile
- temperature, heat index, WBGT above 85th / 90th / 95th percentiles
Days mean temperature index above Standard Deviation(s) of mean daily temperature climate normal
- temperature, heat index, WBGT above 1 or 2 standard deviations of daily normal
Days maximum temperature index above Standard Deviation(s) of max daily temperature climate normal
- temperature, heat index, WBGT above 1 or 2 standard deviations of maximum daily normal

## knitr
## knitr
## tidyverse
## readxl
## lubridate
## viridis
## knitr
## kableExtra
## zoo
## purrr
## furrr
## table1
## furniture
## kableExtra
## visdat
## naniar
## modelr
## lme4
## ICC
## dotwhisker
## broom
## car
## reshape2

Table of HSI Ambulatory Rates

Ambulatory rates (per 1,000 persons per year) of any heat stress illness type (Army personnel)

## knitr
## knitr
## tidyverse
## readxl
## lubridate
## viridis
## knitr
## kableExtra
## zoo
## purrr
## furrr
## table1
## furniture
## kableExtra
## visdat
## naniar
## modelr
## lme4
## ICC
## dotwhisker
## broom
## car
## reshape2
## dplyr
## knitr

Plots of HSI Ambulatory Rates

## knitr
## knitr
## tidyverse
## readxl
## lubridate
## viridis
## knitr
## kableExtra
## zoo
## purrr
## furrr
## table1
## furniture
## kableExtra
## visdat
## naniar
## modelr
## lme4
## ICC
## dotwhisker
## broom
## car
## reshape2
## dplyr
## knitr
## ggridges
## knitr
## knitr
## tidyverse
## readxl
## lubridate
## viridis
## knitr
## kableExtra
## zoo
## purrr
## furrr
## table1
## furniture
## kableExtra
## visdat
## naniar
## modelr
## lme4
## ICC
## dotwhisker
## broom
## car
## reshape2
## dplyr
## knitr
## ggridges
## dplyr

Scatterplots of Temperature/Heat Index and HSI Rates

Each point represents a year from 1997 - 2018.
The back line is a linear regression and the blue curve is a loess smoothed conditional means curve.

## knitr
## knitr
## tidyverse
## readxl
## lubridate
## viridis
## knitr
## kableExtra
## zoo
## purrr
## furrr
## table1
## furniture
## kableExtra
## visdat
## naniar
## modelr
## lme4
## ICC
## dotwhisker
## broom
## car
## reshape2
## dplyr
## knitr
## ggridges
## dplyr

Linear models

## knitr
## knitr
## tidyverse
## readxl
## lubridate
## viridis
## knitr
## kableExtra
## zoo
## purrr
## furrr
## table1
## furniture
## kableExtra
## visdat
## naniar
## modelr
## lme4
## ICC
## dotwhisker
## broom
## car
## reshape2
## dplyr
## knitr
## ggridges
## dplyr
## broom
## broom

Linear Models Summary Table

All models, sorted by R2

## knitr
## knitr
## tidyverse
## readxl
## lubridate
## viridis
## knitr
## kableExtra
## zoo
## purrr
## furrr
## table1
## furniture
## kableExtra
## visdat
## naniar
## modelr
## lme4
## ICC
## dotwhisker
## broom
## car
## reshape2
## dplyr
## knitr
## ggridges
## dplyr
## broom
## broom
## dplyr
## knitr
## dplyr
## knitr

Discussion
Best fit linear models: the annual “temperature/heat” index describes much of the variability in heat stress illness rates.

The installations with the best fit models are Fort Campbell, KY, Fort Benning, GA, and Fort Jackson, SC. These are all sites with high rates of HSIs. Fort Bragg, NC is an exception that has high HSI rates, but is not listed high among best fit linear models.

Best fit indices (R2 above 40%):

  • days_wbgt_max_gt85pct (relative)
  • days_wbgt_max_gt1sd (relative)
  • days_tmp_gt85pct (relative)
  • days_wbgt_gt82
  • hours_wbgt_gt82
  • days_wbgt_gt85
  • days_hi_gt80
  • days_wbgt_max_gt85pct (relative)
  • days_heat_index_gt85pct (relative)
  • days_heat_index_max_gt1sd (relative)
  • days_wbgt_max_gt90pct (relative)
  • days_wbgt_max_gt2sd_may_sep (relative)
  • days_heat_index_max_gt90pct (relative)
  • wbgt_f_max_anomaly_mean_may_sep (relative)
  • days_tmp_gt90pct_may_sep (relative)
  • days_heat_index_gt90pct_may_sep (relative)
  • days_wbgt_max_gt1sd_may_sep (relative)

The top models are relative indices (based on 1990-2018 climatology), aggregated over the entire calendar year.

WBGT models generally fit better than Heat Index or temperature.

Linear Models Summary, by Installation

## knitr
## knitr
## tidyverse
## readxl
## lubridate
## viridis
## knitr
## kableExtra
## zoo
## purrr
## furrr
## table1
## furniture
## kableExtra
## visdat
## naniar
## modelr
## lme4
## ICC
## dotwhisker
## broom
## car
## reshape2
## dplyr
## knitr
## ggridges
## dplyr
## broom
## broom
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr

Note: NA’s are present for models where the conditions were not met (e.g. no days with HI above 125°F, no days with mean temp above 95th percentile).

Installations with poorly fitting models (all R2 below 25%) are:
Fort Riley, KS (highest R2 = 11.8%), Fort Bragg, NC (17.0%), Fort Leonard Wood, MO (19.0%), and Fort Hood, TX (22.8%).

A WBGT-based model fit best for 6 installations, and a Heat Index-based model fit best for 5 (Hood, Jackson, Leonard Wood, Stewart).

Include year in model.

## knitr
## knitr
## tidyverse
## readxl
## lubridate
## viridis
## knitr
## kableExtra
## zoo
## purrr
## furrr
## table1
## furniture
## kableExtra
## visdat
## naniar
## modelr
## lme4
## ICC
## dotwhisker
## broom
## car
## reshape2
## dplyr
## knitr
## ggridges
## dplyr
## broom
## broom
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr

Selected indices (ambulatory)

Based on fits of linear models of annual heat illness ambulatory rates at 11 installations, we selected 10 indices out of 86 tested. Both absolute and relative indices are included. These models generally represent thresholds at the lower end of the heat intensity scales - rather than extreme heat days. 9 of the 10 are based on the full calendar year, rather than just the hot season.

Further analysis: will hospitalizations be closer tied to more extreme heat indices?

Absolute days_wbgt_gt82
hours_wbgt_gt82
days_wbgt_gt85
days_hi_gt80

Relative days_tmp_gt85pct
days_wbgt_max_gt85pct
days_wbgt_max_gt1sd days_heat_index_gt85pct days_wbgt_max_gt90pct
days_tmp_gt90pct_may_sep

Full / Pooled Model

Ambulatory rates, pooled: unweighted by base populations, no indicator variables for installation

## knitr
## knitr
## tidyverse
## readxl
## lubridate
## viridis
## knitr
## kableExtra
## zoo
## purrr
## furrr
## table1
## furniture
## kableExtra
## visdat
## naniar
## modelr
## lme4
## ICC
## dotwhisker
## broom
## car
## reshape2
## dplyr
## knitr
## ggridges
## dplyr
## broom
## broom
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr

Poor fits of pooled, unweighted ambulatory annual rates of heat illness

Pooled linear models, weighted by population

Hypothesis: poor fits due to weighting of Fort Bragg and Fort hood, which had poor fits in separate installation models.

## knitr
## knitr
## tidyverse
## readxl
## lubridate
## viridis
## knitr
## kableExtra
## zoo
## purrr
## furrr
## table1
## furniture
## kableExtra
## visdat
## naniar
## modelr
## lme4
## ICC
## dotwhisker
## broom
## car
## reshape2
## dplyr
## knitr
## ggridges
## dplyr
## broom
## broom
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## dplyr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr

Pooled model with year

Nest by index of heat

## knitr
## knitr
## tidyverse
## readxl
## lubridate
## viridis
## knitr
## kableExtra
## zoo
## purrr
## furrr
## table1
## furniture
## kableExtra
## visdat
## naniar
## modelr
## lme4
## ICC
## dotwhisker
## broom
## car
## reshape2
## dplyr
## knitr
## ggridges
## dplyr
## broom
## broom
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## dplyr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr

Intra-class correlation (ICC)

How much of the variability is explained by the variability across installations.

## knitr
## knitr
## tidyverse
## readxl
## lubridate
## viridis
## knitr
## kableExtra
## zoo
## purrr
## furrr
## table1
## furniture
## kableExtra
## visdat
## naniar
## modelr
## lme4
## ICC
## dotwhisker
## broom
## car
## reshape2
## dplyr
## knitr
## ggridges
## dplyr
## broom
## broom
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## dplyr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## ICC
## ICC

Exposure index: ICC > 0.75
“excellent reproducibility”

Outcome (ambulatory rate): ICC approx. 0.50
“fair to good reproducibility”

Fixed intercepts

Within subject effect estimate: models estimate intercept for each installation.

## knitr
## knitr
## tidyverse
## readxl
## lubridate
## viridis
## knitr
## kableExtra
## zoo
## purrr
## furrr
## table1
## furniture
## kableExtra
## visdat
## naniar
## modelr
## lme4
## ICC
## dotwhisker
## broom
## car
## reshape2
## dplyr
## knitr
## ggridges
## dplyr
## broom
## broom
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## dplyr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## ICC
## ICC
## broom
## dplyr
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dotwhisker

Note: without including year, R2 approx. 52-55%; with year R^2 62-63%

With fixed intercepts added, there is very little difference in model fits based on index selection.

Random intercepts and slopes

Assume independence across subjects i but not within subjects j. Account for within subject correlation with a random term.

(1|installation) designates intercepts only by random factor
(value|installation)designates slopes only by random factor
(1 + value|installation) designates intercepts and slopes by random factor

## knitr
## knitr
## tidyverse
## readxl
## lubridate
## viridis
## knitr
## kableExtra
## zoo
## purrr
## furrr
## table1
## furniture
## kableExtra
## visdat
## naniar
## modelr
## lme4
## ICC
## dotwhisker
## broom
## car
## reshape2
## dplyr
## knitr
## ggridges
## dplyr
## broom
## broom
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## dplyr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## ICC
## ICC
## broom
## dplyr
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dotwhisker
## broom
## car
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## dotwhisker

Random intercepts and slopes

## knitr
## knitr
## tidyverse
## readxl
## lubridate
## viridis
## knitr
## kableExtra
## zoo
## purrr
## furrr
## table1
## furniture
## kableExtra
## visdat
## naniar
## modelr
## lme4
## ICC
## dotwhisker
## broom
## car
## reshape2
## dplyr
## knitr
## ggridges
## dplyr
## broom
## broom
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## dplyr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## ICC
## ICC
## broom
## dplyr
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dotwhisker
## broom
## car
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## dotwhisker
## broom
## car
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## reshape2
## reshape2

Random intercept only models

## knitr
## knitr
## tidyverse
## readxl
## lubridate
## viridis
## knitr
## kableExtra
## zoo
## purrr
## furrr
## table1
## furniture
## kableExtra
## visdat
## naniar
## modelr
## lme4
## ICC
## dotwhisker
## broom
## car
## reshape2
## dplyr
## knitr
## ggridges
## dplyr
## broom
## broom
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## dplyr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## ICC
## ICC
## broom
## dplyr
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dotwhisker
## broom
## car
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## dotwhisker
## broom
## car
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## reshape2
## reshape2
## broom
## car
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## dplyr
## dplyr
## dotwhisker

Random intercept only - without year

## knitr
## knitr
## tidyverse
## readxl
## lubridate
## viridis
## knitr
## kableExtra
## zoo
## purrr
## furrr
## table1
## furniture
## kableExtra
## visdat
## naniar
## modelr
## lme4
## ICC
## dotwhisker
## broom
## car
## reshape2
## dplyr
## knitr
## ggridges
## dplyr
## broom
## broom
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## dplyr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## broom
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## ICC
## ICC
## broom
## dplyr
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dotwhisker
## broom
## car
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## dotwhisker
## broom
## car
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## reshape2
## reshape2
## broom
## car
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## dplyr
## dplyr
## dotwhisker
## broom
## car
## dplyr
## dplyr
## knitr
## dplyr
## knitr
## dplyr
## dplyr
## dplyr
## dotwhisker